WELL-SAMPLED FAR-INFRARED SPECTRAL ENERGY DISTRIBUTIONS OF z ∼ 2 GALAXIES: EVIDENCE FOR SCALED UP COOL GALAXIES

In the top panels of Figure 1, we plot the observed U → 870 μm flux densities of the galaxies. The 24 μm → 870 μm flux densities, which are from dust emission are plotted as black points and the U→8 μm flux densities, which are from stellar photospheres are plotted as blue points. For reference, the best-fit Bruzual & Charlot (2003, hereafter BC03) model to the U→8 μm data from Muzzin et al. (2009) has also been plotted as the solid black line. Figure 1 shows that there is an unambiguous peak in the FIR SEDs of the galaxies at ∼300 μm, a clear signature of the redshifted thermal emission from dust.

Zoom In Zoom Out Reset image size

We fit the 24 μm → 870 μm data using the models of Chary & Elbaz (2001, hereafter CE01). The CE01 models are a grid of 105 synthetic SEDs that were constructed to reproduce the correlations between the observed MIR and FIR fluxes of local galaxies. Each model SED is associated with an L_IR ranging between 3 × 10⁸ L_☉ and 6 × 10¹³ L_☉. We fit the 24 μm → 870 μm data using each of the 105 template SEDs using a χ² minimization method where the amplitude of the SED is the only free parameter.

The galaxy ECDFS-4511 is the most FIR luminous of the two galaxies and is well fit by the SED template of a ULIRG. In the CE01 models that template is associated with an L_IR = 1.1 × 10¹² L_☉; however, ECDFS-4511 has L_IR = 8.9^+1.1_−1.4 × 10¹² L_☉, eight times larger than the luminosity of the local template.

The second galaxy, ECDFS-12514, is three times brighter at 24 μm than ECDFS-4511; however, it has a lower L_IR. The best-fit template is an LIRG template with strong PAH features and is associated with L_IR = 1.2 × 10¹¹ L_☉. The L_IR of ECDFS-12514 is 5.7^+1.2_−0.1 × 10¹² L_☉, almost 50 times more luminous than the local template. If the local template is correct, the large 24 μm flux from ECDFS-12514 likely arises from strong rest-frame 7.8 μm PAH emission redshifted into the observed 24 μm bandpass.

The key result in Figure 1 is that although the z= 2 galaxies have FIR SEDs that are similar to local FIR-luminous galaxies, their luminosities are roughly an order of magnitude larger. It appears that both galaxies are strongly "scaled up" versions of much lower luminosity galaxies in the local universe. The dot-dashed line on the χ² grids in Figure 1 shows the χ² of the fit to the local template that is associated with the best-fit L_IR of ECDFS-4511 and ECDFS-12514. Those χ²s are significantly worse than the best-fit χ² and show that the SEDs of the z∼ 2 galaxies are inconsistent with having the same SED shape of local templates of similar luminosity at >25σ.

Direct constraints on the cool dust SED allow us to address the issue of how well L_IR and SFR(L_IR) can be determined for z∼ 2 galaxies using 24 μm data alone. The most common practice for determining L_IR from 24 μm photometry is to select the local template with an L_IR that would match the observed 24 μm flux if redshifted to the redshift of the source galaxy. In Figure 1, we plot the CE01 template that would be selected based on using that method as the red dotted line.

The extraordinary 24 μm flux of both galaxies requires that they use the most luminous templates in the CE01 library. The implied L_IR for ECDFS-4511 and ECDFS-12514 is 2.8 × 10¹³ L_☉ and 3.8 × 10¹³ L_☉, respectively. These templates overestimate the L_IR from the full SED fit by factors of 3.0 and 6.2, respectively, and it is clear from Figure 1 that they cannot simultaneously fit both the 24 μm flux densities and the FIR flux densities. If these galaxies are representative of most FIR-luminous galaxies in the distant universe, it suggests that we may be systematically overestimating total SFRs by similar factors.

Similar results have also been found by Papovich et al. (2007) and Murphy et al. (2009). Papovich et al. (2007) compared L_IR determined from 24 μm data to that determined using the combined 24 μm, 70 μm, and 160 μm photometry and found that for the most luminous galaxies (S₂₄ > 250 μJy), the 24 μm photometry overestimated the L_IR by factors of 2–10. Using a sample of 14 MIR-luminous galaxies at 1.4 <z< 2.6 with Spitzer IRS spectra and SCUBA 850 μm photometry, Murphy et al. (2009) found that 24 μm fluxes combined with local templates systematically overestimated the L_IR by a factor of ∼5.

The FIR-derived SFRs can also be compared with the dust-corrected Hα SFRs for these galaxies. Hα equivalent widths (EWs) were measured for both galaxies using moderate-resolution NIR spectra from SINFONI by Kriek et al. (2007). We convert their EW(Hα) to an Hα line flux by multiplying by the continuum flux at 6563 Å estimated using the BC03 SED model fits from Muzzin et al. (2009).

We assess the extinction for the Hα flux using both the fit to the stellar SED (U→8 μm) and the Balmer decrement. For the SED fits, we use the Muzzin et al. (2009) best-fit value of A_v, which is derived from fitting BC03 models using solar metallicity and the Calzetti et al. (2000) dust law (hereafter $A_{v,\rm SED}$). For the A_v from the Balmer decrement (hereafter $A_{v,\rm Balmer}$), we assume the Calzetti et al. (2000) dust law and an intrinsic ratio of Hα/Hβ = 2.86. Although both galaxies have high signal-to-noise ratio (S/N) Hα detections in the SINFONI spectra, neither has a secure Hβ detection. Therefore, instead of a true Balmer decrement, we calculate 1σ lower limits⁹ using the 1σ upper limits on Hβ determined by Kriek et al. (2007). The Hβ upper limits have been corrected for absorption by the stellar population using a BC03 SED fit to the broadband photometry and the NIR spectroscopy.

In Figure 2, we plot $A_{v,\rm Balmer}$ versus $A_{v,\rm SED}$ for the two galaxies as the red lower limits. For comparison, measurements of the nine other z∼ 2 galaxies in Kriek et al. (2007) and the galaxy from Kriek et al. (2009b) are also plotted. Two of these galaxies (IDs 1030-807 and CDFS-6202) have secure Hβ detections, the remainder are plotted as 1σ lower limits. The one-to-one relation is shown as the dotted line.

Zoom In Zoom Out Reset image size

The two galaxies with Hβ detections do not agree well with the one-to-one relation. The lower limits for approximately half the galaxies are also inconsistent with the one-to-one relation. A similar disagreement among A_v measurements was found for local galaxies by Calzetti (1997). She showed that this difference can be understood as greater extinction toward the star-forming regions, where the Hα originates, than to the overall stellar population. The Calzetti (1997) relation for local galaxies is E(B − V)_stars = (0.44 ± 0.03)E(B − V)_nebular. In Figure 2, we plot that relation converted to A_v assuming R_v = 4.05 (Calzetti et al. 2000) as the solid line. The Calzetti (1997) relation seems to provide a better description of the overall population of z∼ 2 star-forming galaxies, albeit that most galaxies only have lower limits.

Given that there are large uncertainties in the $A_{v,\rm Balmer}$ and that $A_{v,\rm SED}$ is relatively well-determined, we have chosen to dust correct the Hα flux using the $A_{v,\rm SED}$ but using the E(B − V)_stars = (0.44 ± 0.03)E(B − V)_nebular relation. We convert this to an SFR using the Kennicutt (1998) relation, SFR(Hα) = 0.57 · L(Hα)/1.26 × 10⁻⁴¹ erg s⁻¹, where the factor of 0.57 has been added to convert from a Salpeter initial mass function (IMF) to a Chabrier IMF. Applying these conversions, we find that ECDFS-4511 and ECDFS-12514 have SFR(Hα) = 620⁺²⁶⁰₋₁₄₀ M_☉ yr⁻¹ and 1210⁺⁷⁵⁰₋₄₀ M_☉ yr⁻¹, respectively. We note that the majority of the error budget for these measurements comes from the uncertainty in the extinction correction, not from random errors in the measurement of the Hα line flux.

We convert the L_IR to an SFR using the Kennicutt (1998) relation, SFR(L_IR) = 0.57 · L_IR/5.8 × 10⁹ L_☉, where again the factor of 0.57 is to convert to a Chabrier IMF. We derive SFR(L_IR) = 870⁺¹⁵⁰₋₁₄₀ M_☉ yr⁻¹ and 560⁺¹²⁰₋₁₀ M_☉ yr⁻¹ for ECDFS-4511 and ECDFS-12514, respectively.

Comparing the two SFR indicators shows that they agree to within roughly a factor of two for both galaxies. Such an agreement is good considering that there are still systematic uncertainties in the conversion of both L_IR and L(Hα) to an SFR that are of that order (Kennicutt et al. 2009).

It is worth noting that the agreement in SFRs only occurs because L_IR is determined from the full FIR SED fits, and because the extra extinction correction has been applied to the L(Hα). If we do not use the L_IR from the FIR SED fit and instead use the SFR(24 μm), we find that the IR SFR is larger than the SFR(Hα) by factors of 4.8 and 3.4 for ECDFS-4511 and ECDFS-12514, respectively. Likewise, if we extinction correct the L(Hα) using the $A_{v,\rm SED}$ without extra extinction, we find that the SFR(L_IR) is larger than the SFR(Hα) by a factor of 5.4 and 2.5; however, this time because the SFR(Hα) is underestimated. If we compare the SFR(24 μm) to the SFR(Hα)_SED, the SFR(24 μm) is larger by factors of 18.7 and 18.6.

It is remarkable that despite an L_IR ∼ 10¹³ L_☉, both of the BLAST-detected z∼ 2 galaxies are better fit with local templates that have PAH features and a cool "cold" dust bump. In the nearby universe, galaxies with L_IR ∼ 10¹³ L_☉ tend to have both the warmest "cold" dust (T_dust ∼ 10–70 K, emission peaking in the FIR) and significant amounts of "hot" dust (T_dust ∼ 70–500 K, emission peaking in the MIR). Although both of these dust components can come from deeply embedded star formation (e.g., Tran et al. 2001), it is thought that the dominant radiation source in most, if not all, nearby galaxies with L ∼ 10¹³ L_☉ is an AGN (e.g., Genzel et al. 1998; Lutz et al. 1998).

By contrast, the best-fit templates of the z∼ 2 galaxies are much cooler ULIRG and LIRG templates which are most frequently associated with star-forming galaxies in the local universe (e.g., Armus 2009). These templates also tend to have stronger PAH features that are more prevalent because the "hot" dust component from the AGN no longer dominates the MIR emission.

The first claims that IR-luminous galaxies at z∼ 2 may be scaled up versions of lower luminosity cooler galaxies came from 24 μm observations of submillimeter-selected galaxies (SMGs). Pope et al. (2006) showed that the SEDs of SMGs peak at longer wavelengths than expected based on an extrapolation of their 24 μm flux using local templates, arguing that SMGs are cooler at a given luminosity than the local templates. Subsequent MIR spectroscopic observations of SMGs from Spitzer by Lutz et al. (2005), Pope et al. (2008), and Menéndez-Delmestre et al. (2009) all found strong PAH features and relatively weak MIR continuum in the SMGs suggesting much less "hot" dust than is usually found in local galaxies of comparable L_IR. Similarly, Papovich et al. (2009) found a weaker than expected rest-frame 24 μm luminosity for a lensed SMG, arguing for the lack of a "hot" dust component, and hence a scaled up cool galaxy.

With access to the FIR we can determine a dust temperature (T_d) independent of the MIR SED and compare this to the local T_d–L_IR relation. We fit the 160 μm → 870 μm data (50 μm → 300 μm rest frame) to modified blackbody curves of the form S(ν) = Aν^βB(ν,T_d), where A is a normalization, B is the Planck function, and β accounts for frequency-dependent dust emissivity. Even with detections in numerous bands the S/N of the data is too low to constrain both β and T_d simultaneously, so we have assumed β = 1.5, a typical value for submillimeter galaxies. The 70 μm points are omitted in the fitting because they correspond to rest frame 23 μm and are too far into the MIR to include in a single-component blackbody fit. We note that including the 70 μm data actually produces identical best-fit temperatures; however, the χ² of fits are substantially larger because no single-temperature model can describe the 23 μm → 300 μm rest frame simultaneously.

We find dust temperatures of 40⁺²₋₁ K and 41⁺⁵₋₇ K for ECDFS-4511 and ECDFS-12514, respectively. These correspond to rest-frame monochromatic IRAS color measurements, C ≡ log₁₀(f_60 μm/f_100 μm), of −0.05^+0.05_−0.03 and −0.03^+0.09_−0.20, respectively. Comparing these to the C–L_IR relation measured by Chapin et al. (2009) in the nearby universe using IRAS shows that these galaxies have temperatures typical of nearby galaxies with L_IR = 10¹¹–10¹² L_☉. They are also similar to the T_d of local LIRGS/ULIRGS measured by Clements et al. (2010). Therefore, with the combined Spitzer, BLAST, and APEX data we can now confirm that some massive galaxies at z∼ 2 not only have PAH-dominated MIR SEDs similar to lower luminosity local galaxies, but they also have "cold" dust SEDs consistent with being scaled up versions of lower luminosity galaxies. This suggests that star formation may be occurring in a different environment within galaxies at z∼ 2 compared to z∼ 0.

Local galaxies with SFRs >100 M_☉ yr⁻¹ tend to have compact nuclear starbursts which are the cause of the warmer dust temperatures (e.g., Sanders & Mirabel 1996). The z∼ 2 galaxies have SFRs ∼ 1000 M_☉ yr⁻¹, but similar dust temperatures which suggests that the star formation may be occurring in multiple large starburst regions throughout the galaxy, rather than concentrated in a single compact starburst which would presumably have a substantially hotter dust temperature. Indeed, a lensed z>2 galaxy with precisely such characteristics has recently been observed by Swinbank et al. (2010a) using interferometric submillimeter observations.

Kriek et al. (2009a) presented SINFONI-IFU Hα spectra of both these galaxies. Their maps show that the Hα emission is resolved on scales of a few kpc and is also consistent with the extended star formation hypothesis. The greater availability of gas at high redshift may permit more spatially extended star formation at a high rate, something that is uncommon in local galaxies.

In the inset of Figure 1, we show 3'' × 3'' NIC2 F160W images of the galaxies from Kriek et al. (2009a). At z∼ 2, these correspond to rest-frame V-band morphologies and span an angular size of ∼25 kpc. Like most of the high redshift ULIRGs and HLIRGs previously observed with Hubble Space Telescope (e.g., Chapman et al. 2003; Dasyra et al. 2008; Swinbank et al. 2010b) the galaxies clearly have disturbed morphologies. Kriek et al. (2009a) measure effective radii (R_e) of 5 kpc and 3 kpc for ECDFS-4511 and ECDFS-12514, respectively, which means that both galaxies are comparable in size to local galaxies of similar mass. Rest-frame optical morphologies make it difficult to identify the location of the star formation (which is presumably highly obscured); however, the disturbed nature of both galaxies shows that these are complex systems, possibly "stream-fed" galaxies (Dekel et al. 2009) or mergers. Unlike the FIR SEDs, the morphologies do not appear to be scaled up versions of local star-forming disk galaxies.

Despite SEDs that are associated with local star-forming galaxies, part of the MIR and FIR emission in both galaxies almost certainly comes from an AGN. Using emission line ratios and the BPT diagram (Baldwin et al. 1981), Kriek et al. (2007) found that both galaxies were in the region of the diagram that Kewley et al. (2006) considered to have "composite" spectra with contributions from both star formation and an AGN.

ECDFS-4511 is detected in the 2 Ms Chandra map of the CDFS (Luo et al. 2008) and has an L_x ∼ 2 × 10⁴⁴ erg s⁻¹, implying it hosts a Seyfert-luminosity AGN. ECDFS-12514 is located in the 250 ks map of the ECDFS (Virani et al. 2006), but is undetected. Although undetected in the X-rays, ECDFS-12514 is formally classified as an IRAC power-law galaxy (F_ν ∝ ν^α, α < −0.5; e.g., Donley et al. 2007; see Figure 1) which makes it a candidate for an obscured AGN.

Without more data we cannot constrain the precise fraction of the MIR and FIR light that comes from the AGN; however, the consistency of the MIR/FIR SED with local templates that are star formation dominated and the "composite" emission line ratios suggest that young stars are probably the dominant energy source heating the dust. Pope et al. (2008) and Menéndez-Delmestre et al. (2009) showed that the MIR SEDs of most SMGs (the BLAST-detected galaxies are formally SMGs) are still dominated by PAHs and that AGNs only contribute ∼30% of the total MIR emission. Putting the evidence together the most plausible scenario is that both galaxies host a moderate-luminosity AGN, but the AGN does not dominate the MIR/FIR SED. Nevertheless, we note that both our SFR(L_IR) and SFR(Hα) should be formally considered upper limits as the AGN will contribute to both of these.

The results of our FIR SED fitting show that caution is required when determining and interpreting the SFRs of z∼ 2 galaxies based on limited data. Using only 24 μm photometry to infer SFRs at z∼ 2 may work well for lower luminosity or less obscured galaxies (e.g., Papovich et al. 2007; Reddy et al. 2010); however, the local templates can overestimate the SFR(L_IR) of the most FIR-luminous galaxies by as much as a factor of ∼6 (see also Papovich et al. 2007; Murphy et al. 2009). Likewise, the Balmer decrement measurements for these galaxies seem to indicate extra extinction toward the star-forming regions. It appears that on average this relation is similar to that suggested by Calzetti (1997); however, the lack of secure Hβ detections in our sample precludes quantitative statements. To correctly determine the SFR(Hα) of individual galaxies, rather than ensemble averages requires high-quality Balmer decrement measurements.

The extra extinction toward the star-forming regions suggested by the Balmer decrements also indicates that the UV SFRs for some of these IR-luminous galaxies may not be reliable because the A_v is large enough so that parts of the star-forming regions are optically thick. The SFRs estimated from a fit to the stellar populations (SFR(SED); effectively a dust-corrected UV SFR) by Muzzin et al. (2009) for ECDFS-4511 and ECDFS-12514 are 168⁺⁴¹₋₆₃ M_☉ yr⁻¹ and 145⁺¹⁶⁰₋₀ M_☉ yr⁻¹ and underestimate the SFR(L_IR) of the galaxies by factors of 5.2 and 3.5, respectively. If we compare the SFR(SED) to the SFR(24 μm) as in, e.g., Daddi et al. (2007b), we find that these galaxies have MIR SFRs 18–25 times greater than the UV SFRs and are extremely strong "MIR-excess" galaxies.

The purpose of the Daddi et al. (2007b) classification was to identify candidate obscured AGN, and given that it is likely that both galaxies contain an AGN, it appears that method successfully identifies such sources. However, MIR excesses of factors of 18–25 would suggest that the AGN is the dominant MIR energy source for both galaxies, something that is unlikely given the shape of the MIR/FIR SEDs and the emission line ratios. This calls into question whether galaxies with weaker inferred MIR excesses, such those near the cutoff of SFR(UV_corr)/SFR(MIR+UV) ∼ 3 suggested by Daddi et al. (2007b) have a significant contribution from an AGN.

Our results suggest that a substantial population of massive z∼ 2 galaxies are "scaled up" versions of lower luminosity galaxies. Clearly, such a population needs to be understood within the context of modern galaxy formation and evolution models. Recently, Hopkins et al. (2010) have used their merger-driven evolutionary model to predict the IR luminosity functions of galaxies and quasars between 0 <z< 6. In their model, they argue that there is a threshold in L_IR above which disks, merger-driven star formation, and obscured AGN are the dominant IR energy source of galaxies. At z = 0, these limits are log(L_IR) ⩾ 11.5 for merger-driven star formation and log(L_IR) ⩾ 12.5 for obscured AGN. Based on the evolution of the merger rate and gas fraction to z∼ 2, they suggest that these thresholds evolve in log(L_IR) to log(L_IR) ⩾ 12.5 for merger-driven star formation and log(L_IR) ⩾ 13.0 for obscured AGN. With log(L_IR) ∼ 13.0, the BLAST galaxies lie at an L_IR that is near the suggested transition point between merger-driven star formation dominated or AGN dominated.

Interestingly, the FIR SEDs of these galaxies do not look like those of the AGN-dominated HLIRGs detected by Infrared Space Observatory (Genzel et al. 1998), or of the average nearby ULIRG, which are mostly merger-driven starbursts (e.g., Sanders et al. 1988). Instead, these galaxies look like scaled up versions of even lower luminosity systems, which are not universally associated with mergers or AGNs. Taken at face value, it appears that they do not fit well within the Hopkins et al. (2010) description.

On the other hand, the NIC2 morphologies of the galaxies are clearly disturbed enough to be considered candidates for undergoing mergers (Kriek et al. 2009b). If these galaxies are merger-driven starbursts, it may be that they are missing the centrally concentrated starburst which produces the much hotter dust seen in these galaxies in the local universe. As Hopkins et al. (2010) point out, the efficiency of a merger at funneling gas and stars into the core of the galaxy is proportional to the gas fraction, with higher gas fraction mergers being less efficient. If both galaxies have high gas fractions, most of the star formation could be occurring on much larger physical scales than similar mass starbursts in the local universe, hence the cool dust SED.

Galaxies with similar stellar masses (M ∼ 10¹¹ M_☉) and gas fractions as high as 80% have been observed in the distant universe (e.g., Tacconi et al. 2010) and could be analogs of the BLAST-detected galaxies. If that is the case, it begs the question of what are the descendants of extended, massive, gas-rich star-forming galaxies at z∼ 2? Local spheroids seem like an obvious candidate; however, some fraction of the progenitors of those systems are clearly small and quiescent at z∼ 2 (e.g., van Dokkum et al. 2010; Hopkins et al. 2010; Bezanson et al. 2009; van der Wel et al. 2008). If the large and active BLAST sources are also progenitors of massive local spheroids, it suggests that there may be very different evolutionary paths which arrive at the same galaxies.

These galaxies also appear to be consistent with the "stream-fed" galaxies proposed by Dekel et al. (2009) and the scaled up "turbulent disks" discussed by Genzel et al. (2008). In their model, Dekel et al. (2009) suggest that massive galaxies may grow quickly through rapid cold gas accretion and clumpy star formation which funnels stars into a central bulge. Given that these cold flows often separate into multiple large star-forming clumps, it might be expected that these clumps produce cooler dust temperatures at fixed SFR than centrally concentrated starbursts.

The SFRs of the BLAST galaxies are higher than the 100–200 M_☉ yr⁻¹ predicted for "stream-fed" galaxies by Dekel et al. (2009). The BLAST galaxies are the most IR luminous of the K-selected sample of Kriek et al. (2008), so if they are "stream-fed" galaxies, they may just represent the most active tail of the distribution.

High-quality kinematic data could potentially distinguish between merger-dominated or accretion-dominated systems. These galaxies do have SINFONI IFU observations; however, these data lack the S/N to reliably determine V/σ. Therefore, at present based on the MIR/FIR properties and morphologies, we cannot distinguish between evolutionary models such as suggested by both Hopkins et al. (2010) and Dekel et al. (2009).